The world of artificial intelligence (AI) has been rocked by a recent statement from Meta, the technology company formerly known as Facebook. In an unusual move, Meta said that it was now working on creating its own bespoke processor that would only be used to run AI models. As the company is not normally associated with providing cloud computing services like industry titans Google and Microsoft, the move represents a significant departure from Meta’s usual focus.
Unexpectedly, Meta has managed to keep its internal data center chip project secret until now, never releasing details to the public. Meta’s management has decided to break the silence and share this great news with the public due to the enormous curiosity of the global tech community.
Also read: Facebook’s New Direction: MetAI – Focusing on Artificial Intelligence for a Transformed Future
The Meta Training and Inference Accelerator (MTIA) chip has one outstanding feature that sets it apart from its competitors. With only 25 watts of power consumption, the MTIA far outperforms processors from well-known manufacturers such as NVIDIA, which generally require far more power.
Meta software developer Joel Coburn explained why the MTIA was created rather than using GPUs for AI inference workloads. Although GPUs have been widely used for these applications, Meta has found that they are not the most effective or economical choice for real-world AI model applications. “GPUs are still expensive to deploy and inefficient for real-world modelling applications,” said Coburn. That is why the MTIA is necessary.
Alongside the MTIA, Meta also announced the “scalable video processor” (MSVP), another cutting-edge device. Given Meta’s massive daily video processing requirements, which total an astounding 4 billion videos, this chip focuses primarily on video processing and transmission while reducing power requirements.
The launch of these specialized chips demonstrates Meta’s commitment to advancing AI technology and solving the industry’s unique problems. Meta aims to maximize the effectiveness and performance of its AI models by creating its own specialized hardware, which could lead to new developments and opportunities across a range of industries. The future development and implementation of these ground-breaking chips are being eagerly anticipated by the IT world, as they have the potential to completely change the AI environment.
